If someone was trying to convince you NOT to start a predictive maintenance program, they would probably tell you things like:
“It is just too expensive to implement; it is way too complicated; you don’t really need it…”
And while those might be valid concerns for some, they definitely aren’t stopping many organizations that are implementing it as we speak.
If you too are considering starting a predictive maintenance program at your facility, read through this guide and learn about all the steps you need to take to ensure successful implementation.
Is It Hard To Start Predictive Maintenance Program?
Well, that depends. Writing a simple SOP can be difficult if you have no idea what you’re doing.
To be fair though, implementing a PdM program is more complicated when compared to starting a preventive maintenance program. How hard this process is going to be will depend on your level of expertise, the complexity of your assets, and the tools you have at your disposal.
We recognized the arguments against predictive maintenance and looked for ways to simplify your transition to PdM. That is why we optimized Limble to be able to communicate with predictive equipment and introduced a modular IoT sensor package that lets you start predictive maintenance for under $1000.
Informing yourself by reading this guide is a step in the right direction and will make the whole process significantly smoother.
Steps For Establishing A PdM Program
Although the benefits of predictive maintenance are numerous, deployment can be challenging. So, before you start, it is important to have a well-defined plan covering the desired business outcome and the project scope. Ideally, it’s advisable to start small, learn and adjust as you go along, then use the insights gained for further expansion.
Below are step-by-step best practices for implementing a successful and sustainable PdM strategy.
Step #1) Identify assets for PdM
First, you need to determine which assets you want to include in your predictive maintenance program. This distinction is important because of two reasons:
- Not all assets deserve to be on a predictive maintenance plan as for some that simply isn’t cost-effective. Certain assets are largely expendable, so they can be placed on basic routine maintenance or allowed to run to failure.
- If this is your first tango with predictive maintenance, you don’t want to complicate things from the start. It is better to focus on a few selected assets and run this as a pilot project.
By checking historic machine records – maybe over a two or three year period – you can identify assets that are most vital for your business processes and would cause a substantial disruption if they unexpectedly failed.
A quick way to identify such assets is to note:
- Assets that demand the most financial and human resources.
- Assets with high repair/replacement costs that also record frequent breakdown incidents.
- Equipment which breakdown limits or halts production or service delivery.
- Assets that do not directly impact production, but the repair costs are considerable and/or take longer to complete.
- Machines that are so sensitive or complex that they require “specialist” attention to get them back online (which often comes along with a big invoice).
Assets that meet the above criteria are the best candidates for the PdM program.
For example, by installing vibration and motor current sensors on major electric motors and placing them on a predictive maintenance program, you can avoid breakdown of sensitive production equipment on your plant floor and avoid profit losses running into thousands of dollars per hour of downtime.
Step #2) Establish the presence of actionable data
Machine records present a valuable and time-saving source of actionable data to help get PdM rolling. Such data offers information about machine behavior that will, to a large extent, determine how the PdM model is designed.
Sources of historical machine data include:
New equipment always comes from the manufacturer with comprehensive manuals and instructions. This information covers important details like what kind of maintenance is required, how often it should be done, how to perform it safely, and so on. After commissioning equipment, the manufacturers’ instructions usually form the basis for starting a maintenance plan on each asset.
In-house historical data
Companies that have been keeping accurate maintenance records have an added advantage here as they can quickly gather historical data for each of the selected equipment. Machine data can be extracted from hard copy maintenance records and charts, your CMMS software, and software from other departments (e.g. procurement and accounting).
For instance, this is how an asset report looks like in Limble:
Looking at metrics like total costs, future estimated costs, and MTBF shows you how often an asset fails which might give you a clue if an asset is worth putting on a predictive maintenance plan.
Leverage staff expertise
Besides documentation, you can get valuable information that’s not available in writing by involving the staff that works with the machines daily like maintenance technicians and machine operators. They already have a working knowledge of each asset and are familiar with many failure patterns. Getting their involvement and buy-in will make it easier to pinpoint the specific problems they face and how the eventual predictive model can help.
Step #3) Analyze failures
At this point, you’ll need to perform analysis on the assets you decided to place on a PdM program. The goal of this analysis is to identify their failure modes. It is important to focus on establishing the following:
- Severity (and effect) of machine failure.
- The frequency of failure.
- The difficulty of identifying the failure.
One of the most accurate methods for calculating this is Failure Mode and Effects Analysis (FMEA). FMEA is a product/process reliability analysis tool used to identify failure affecting a system, prioritize corrective action, and limit the effects of failure. It is used to identify failure modes and estimate the risk of each mode. Failure “modes” are the different events or ways in which an asset can fail.
FMEA is a very detailed and thorough process that involves several worksheets and calculations but in summary, it functions like this:
- Identify the assets and list their normal functions.
- Consider different failures for the asset.
- Identify the effect of each failure (effect on the asset, on people, the surroundings, etc.)
- Rank the severity of each failure (usually from 1 to 10).
- Determine the occurrence of each failure mode.
- Assign a ranking for ease of detecting each failure.
- Multiply severity by occurrence and detection to calculate the risk priority number (RPN) for each failure.
- Assign actions to the high-risk failure modes.
- Review and re-rank RPNs as failure reduces with time.
The result is a prioritized list that guides the implementation team to begin work on developing the failure predictions for the highest risk assets first.
Step #4) Choose and implement condition monitoring techniques and equipment
The predictive maintenance program integrates different types of machine information such as performance data, maintenance history, and design data to make timely decisions about maintenance intervention. To achieve this, it requires specific technologies and real-time equipment condition data to function effectively. This is possible through condition-based monitoring.
Condition-based monitoring is a key step in the process and it works on the assumption that all machines will deteriorate and fail partially or fully at some point. Therefore, the goal is to preempt these failures by placing various monitoring sensors on the assets. From there, the data is collected, analyzed, and used to create predictive failure algorithms which inform your maintenance actions.
There are a wide variety of sensors available including (but not limited to):
- Thermal cameras
- Leak detectors
The most common condition monitoring techniques used to detect these faults are:
- Vibration analysis
- Lubricant analysis
- Infrared thermography
- Ultrasound testing
- Dynamic pressure analysis
- Acoustics testing
- Current and voltage testing
Identifying the failure modes for your critical assets as described in step 3 helps you to choose the appropriate monitoring technique and equipment for each asset. For example, vibration analysis is the most commonly used technique for rotating equipment as it can detect the faults that this category of equipment are prone to, such as roller bearing wear, mechanical looseness, gearbox wear, shaft misalignment, and unbalance.
There are several reputable brands of sensors you could choose, manufactured by companies like Siemens AG, Schneider Electric, Bosch GmbH, ABB, Honeywell, to mention a few.
Step #5) Develop algorithms for making failure predictions
Next is developing the algorithms that will form the basis for predicting each of the failure modes identified with FMEA. This is the core of predictive maintenance and what happens here will get you the equipment monitoring alerts you need.
The predictive system uses both the information coming from condition monitoring sensors and prognostics algorithms to analyze machine data. Let’s take a brief look at how both functions interact:
1) Condition monitoring
The metrics derived from condition monitoring sensors track any behavior changes that indicate the asset is degrading. It does this by clustering similar/acceptable condition parameters (vibration, temperature, noise, etc) together then setting any deviations aside. This way, it performs fault detection by comparing sensor data against previously set markers of faulty conditions.
2) Prognostics algorithms
Prognostics algorithms are used to estimate the remaining time-to-failure (RTTF) or the remaining useful life (RUL) of a machine or component. It can help to forecast when failure will occur by comparing the condition of a machine over a period of time. These algorithms can be established using either modeling, machine learning technology, or by combining both.
Predictive modeling is initially done by a data scientist who creates predictive models. Then, a machine learning platform technology is incorporated to update algorithms steadily, increasing its predictive capabilities with each asset failure incident until unplanned downtime can almost be completely eliminated.
Although this process may take a while to perfect, the final result is an automated system that calculates the RTTF, can generate alerts when machine conditions deviate from established thresholds, and determine when maintenance intervention should happen.
Step #6) Deploy to pilot equipment
You now have detection for monitoring and prediction for prognostics sorted. The final step of implementation is to deploy the technology by integrating it into a few selected pilot assets.
For starters, the algorithms can run on local embedded devices mounted near the monitored assets. This works since the amount of data generated doesn’t go to the extreme for just a few assets. However, as you begin to scale up to include dozens or even hundreds of assets, considerable amounts of time and computing resources would be required to analyze the data generated from large-scale deployment.
Fortunately, the cloud and advances in artificial intelligence provide the platform to quickly collect and analyze machine data in real-time and at incredible speeds. This enables rapid PdM expansion to include more and more machines. It’s therefore advisable to opt for cloud-based integration at that point.
The results from the cloud-based analysis are made available to the end-users in the medium they want but most commonly through email notifications, on dashboards, tweets, and other web applications.
Is Your Facility Ready For A Predictive Maintenance Program?
While PdM represents a very powerful proactive approach to maintenance, it does have a barrier to entry.
Below are the requirements you should be able to check off if you are serious about implementing a predictive maintenance program:
1) Upper management support
Starting a predictive maintenance program is a big step forward for any facility and is definitely not a project you want to run without having strong support from upper management. You will need funding, help from other departments, and possibly even to hire a third-party consultant. You don’t want to be left stranded in the middle of the implementation process.
Setting up a pilot project and laying down realistic expectations can go a long way in getting the necessary approval from the people in charge.
2) Appropriate funding
This builds upon the previous point. As you will need to buy sensor monitoring equipment, invest in a CMMS or other specialized software, possibly hire outside data scientists to help you with creating predictive models, and spend some resources on training your technicians, having a flexible budget is a big plus.
3) Right condition monitoring equipment
The real power of predictive maintenance hides behind 2 simple concepts: the ability to monitor the condition of your assets in real-time and feeding that data into complex algorithms that allow you to perfectly schedule necessary repairs and replacements.
Without appropriate condition-monitoring equipment, the algorithms simply won’t have enough data to do their job properly and give you accurate predictions and alerts.
4) Access to right software and analytics
Arguably the most important requirement alongside CM equipment is ensuring you have the ability to analyze incoming data and develop predictive maintenance models. It is not rare that businesses look for outside help with this part of the process until someone inside their organization is sufficiently trained to take over.
In regards to software requirements, CMMS will be insanely helpful in creating and running your PdM program by helping you schedule and oversee all maintenance activities, from ensuring you have necessary spare parts when you need them, to tracking task progress and maintenance costs. More advanced CMMS solutions can even be configured to automatically create tasks depending on the incoming sensor data. Here is an example of such a task in Limble CMMS when the vibration rate exceeds your predefined limit:
In addition to CMMS, some organizations might also need to look into employing advanced predictive analytics that will help them with the creation of necessary predictive models.
5) Training requirements
While a lot of condition monitoring sensors can be connected to software like CMMS and give you a constant feed of information, some CM equipment might need manual input and monitoring from the side of your maintenance technicians who need to know how to use it.
Additionally, no matter which type of maintenance strategies you were using so far, a move towards predictive maintenance will require some workflow changes on the scope of the whole maintenance department.
This is yet another reason why starting with a pilot project is a good idea – it gives everyone enough time to get familiar with all of the incoming changes.
A Switch To Predictive Maintenance
Managers often like to wait for the right time to switch to predictive maintenance, which is understandable considering it can be a big project. However, if you are going to wait until the stars align and conditions are picture-perfect, you are going to postpone it forever.
If you’re on top of most of the requirements mentioned above, you should be able to begin a pilot project and work on addressing what you’re missing as you go.
For everyone that is serious about starting a PdM program and is interested in our modular IoT sensor package, don’t hesitate to contact us and see how we can help you.